def giveChildren(self, node, includeClassToPredict=False):
"""Gives children to <node>, returns these children"""
leftClass = self.pickAnAttribute(includeClassToPredict)
rightClass = self.pickAnAttribute(includeClassToPredict)
leftNode = GPNode(leftClass, self.pickALimit(leftClass))
rightNode = GPNode(rightClass, self.pickALimit(rightClass))
node.setLeftChild(leftNode)
node.setRightChild(rightNode)
return [leftNode, rightNode]
def addTerminatingNodes(self, currentNodes):
"""Adds terminal nodes to all nodes in <currentNodes>"""
# BE CAREFUL HERE; do I want the final split to have the possibility of predicting the same finalValue?
terminatingOptions = self.categoryLimits[self.classToPredict]
for node in currentNodes:
optionLeft = random.choice(terminatingOptions)
terminatingOptions.remove(optionLeft)
optionRight = random.choice(terminatingOptions)
node.setLeftChild(GPNode(self.classToPredict, optionLeft))
node.setRightChild(GPNode(self.classToPredict, optionRight))
terminatingOptions.append(optionLeft)
def initBalancedTree(self):
"""Creates a decision tree of depth <self.depth>"""
nodesAtCurrentDepth = []
rootClass = self.pickAnAttribute()
self.root = GPNode(rootClass, self.pickALimit(rootClass))
nodesAtCurrentDepth.append(self.root)
depth = 0
# Initialize tree to one level short of depth
while depth < self.maxDepth - 2:
children = []
for node in nodesAtCurrentDepth:
children += self.giveChildren(node)
nodesAtCurrentDepth = children
depth += 1
# Add the last level of nodes, which should be terminating nodes (i.e. the possible predictions)
self.addTerminatingNodes(nodesAtCurrentDepth)
def mutateDeletion(ind):
"""Delete a random node in <ind>'s tree; replace it with a terminal node"""
tree = ind.tree
terminalOptions = tree.categoryLimits[tree.classToPredict]
terminalReplacement = GPNode(tree.classToPredict, random.choice(terminalOptions))
nodeToDelete, parentNode, left = getRandomNode(tree.root, random.randint(0, tree.maxDepth))
if left:
parentNode.setLeftChild(terminalReplacement)
terminalReplacement.setLeftChild(None)
terminalReplacement.setRightChild(None)
else:
parentNode.setRightChild(terminalReplacement)
terminalReplacement.setLeftChild(None)
terminalReplacement.setRightChild(None)
def initUnevenTree(self):
"""Creates a random decision tree that is not necessarily symmetrical or full"""
rootClass = self.pickAnAttribute() # Not the classToPredict
rootValue = self.pickALimit(rootClass)
self.root = GPNode(rootClass, rootValue)
nodesNeedingChildren = []
nodesNeedingChildren.append(self.root)
depth = 0
while depth < self.maxDepth - 2:
allChildren = []
for node in nodesNeedingChildren:
if node.attribute != self.classToPredict:
# Not a terminal node
allChildren += self.giveChildren(
node, includeClassToPredict=True)
nodesNeedingChildren = allChildren
depth += 1
self.removeTerminalNodes(nodesNeedingChildren)
self.addTerminatingNodes(nodesNeedingChildren)
def initBalancedTree(self):
"""Creates a decision tree of depth <self.depth>"""
nodesAtCurrentDepth = []
rootClass = self.pickAnAttribute()
self.root = GPNode(rootClass, self.pickALimit(rootClass))
nodesAtCurrentDepth.append(self.root)
depth = 0
# Initialize tree to one level short of depth
while depth < self.maxDepth - 2:
children = []
for node in nodesAtCurrentDepth:
children += self.giveChildren(node)
nodesAtCurrentDepth = children
depth += 1
# Add the last level of nodes, which should be terminating nodes (i.e. the possible predictions)
self.addTerminatingNodes(nodesAtCurrentDepth)
def initUnevenTree(self):
"""Creates a random decision tree that is not necessarily symmetrical or full"""
rootClass = self.pickAnAttribute() # Not the classToPredict
rootValue = self.pickALimit(rootClass)
self.root = GPNode(rootClass, rootValue)
nodesNeedingChildren = []
nodesNeedingChildren.append(self.root)
depth = 0
while depth < self.maxDepth - 2:
allChildren = []
for node in nodesNeedingChildren:
if node.attribute != self.classToPredict:
# Not a terminal node
allChildren += self.giveChildren(node, includeClassToPredict = True)
nodesNeedingChildren = allChildren
depth += 1
self.removeTerminalNodes(nodesNeedingChildren)
self.addTerminatingNodes(nodesNeedingChildren)
class GPDecisionTree(object):
def __init__(self, allData, categories, classToPredict, categoryLimits, maxDepth = 5, root = None, balanced = False):
"""Takes in parameters, creates an a randomly initialized decision tree"""
self.root = root
self.maxDepth = maxDepth
self.allData = allData # List of Entry objects
self.categories = categories # Doesn't include classToPredict
self.classToPredict = classToPredict
self.allCategories = categories + [self.classToPredict] # Includes classToPredict
self.categoryLimits = categoryLimits
if balanced:
self.initBalancedTree()
else:
self.initUnevenTree()
def initBalancedTree(self):
"""Creates a decision tree of depth <self.depth>"""
nodesAtCurrentDepth = []
rootClass = self.pickAnAttribute()
self.root = GPNode(rootClass, self.pickALimit(rootClass))
nodesAtCurrentDepth.append(self.root)
depth = 0
# Initialize tree to one level short of depth
while depth < self.maxDepth - 2:
children = []
for node in nodesAtCurrentDepth:
children += self.giveChildren(node)
nodesAtCurrentDepth = children
depth += 1
# Add the last level of nodes, which should be terminating nodes (i.e. the possible predictions)
self.addTerminatingNodes(nodesAtCurrentDepth)
def initUnevenTree(self):
"""Creates a random decision tree that is not necessarily symmetrical or full"""
rootClass = self.pickAnAttribute() # Not the classToPredict
rootValue = self.pickALimit(rootClass)
self.root = GPNode(rootClass, rootValue)
nodesNeedingChildren = []
nodesNeedingChildren.append(self.root)
depth = 0
while depth < self.maxDepth - 2:
allChildren = []
for node in nodesNeedingChildren:
if node.attribute != self.classToPredict:
# Not a terminal node
allChildren += self.giveChildren(node, includeClassToPredict = True)
nodesNeedingChildren = allChildren
depth += 1
self.removeTerminalNodes(nodesNeedingChildren)
self.addTerminatingNodes(nodesNeedingChildren)
def removeTerminalNodes(self, listOfNodes):
"""Removes all terminal nodes from <listOfNodes>"""
nodesToRemove = []
for x in listOfNodes:
if x.attribute == self.classToPredict:
nodesToRemove.append(x)
for node in nodesToRemove:
listOfNodes.remove(node)
def giveChildren(self, node, includeClassToPredict = False):
"""Gives children to <node>, returns these children"""
leftClass = self.pickAnAttribute(includeClassToPredict)
rightClass = self.pickAnAttribute(includeClassToPredict)
leftNode = GPNode(leftClass, self.pickALimit(leftClass))
rightNode = GPNode(rightClass, self.pickALimit(rightClass))
node.setLeftChild(leftNode)
node.setRightChild(rightNode)
return [leftNode, rightNode]
def addTerminatingNodes(self, currentNodes):
"""Adds terminal nodes to all nodes in <currentNodes>"""
# BE CAREFUL HERE; do I want the final split to have the possibility of predicting the same finalValue?
terminatingOptions = self.categoryLimits[self.classToPredict]
for node in currentNodes:
optionLeft = random.choice(terminatingOptions)
terminatingOptions.remove(optionLeft)
optionRight = random.choice(terminatingOptions)
node.setLeftChild(GPNode(self.classToPredict, optionLeft))
node.setRightChild(GPNode(self.classToPredict, optionRight))
terminatingOptions.append(optionLeft)
def pickAnAttribute(self, includeClassToPredict = False):
"""Picks a random attribte (each entry in allData contains every attribute as its keys)"""
if includeClassToPredict:
attr = random.choice(self.allCategories)
else:
attr = random.choice(self.categories)
return attr
def pickALimit(self, attribute):
"""Returns a random limit for the given <attribute> from somewhere in its specified values (as maintained by categoryLimits)"""
possibleLimits = self.categoryLimits[attribute]
try:
# Get random number between lower and upper limit
limit = random.uniform(possibleLimits[0], possibleLimits[1])
except TypeError:
# possibleLimits actually has strings, so <attribute> is categorical; return a random choice from the possibleLimits
limit = random.choice(possibleLimits)
return limit
def notTerminatingNode(self, node):
"""Determines whether or not <node> is a terminating node"""
if node.attribute == self.classToPredict:
return False
else:
return True
def evaluate(self, entries):
"""Tests all entries on his decision tree, returns the number classified correctly, and total entries attempted"""
correct = 0
total = len(entries)
for ent in entries:
if self.root.evaluate(ent):
correct += 1
return correct, total
def __str__(self):
"""Root has everything stored"""
return str(self.root)
class GPDecisionTree(object):
def __init__(self,
allData,
categories,
classToPredict,
categoryLimits,
maxDepth=5,
root=None,
balanced=False):
"""Takes in parameters, creates an a randomly initialized decision tree"""
self.root = root
self.maxDepth = maxDepth
self.allData = allData # List of Entry objects
self.categories = categories # Doesn't include classToPredict
self.classToPredict = classToPredict
self.allCategories = categories + [self.classToPredict
] # Includes classToPredict
self.categoryLimits = categoryLimits
if balanced:
self.initBalancedTree()
else:
self.initUnevenTree()
def initBalancedTree(self):
"""Creates a decision tree of depth <self.depth>"""
nodesAtCurrentDepth = []
rootClass = self.pickAnAttribute()
self.root = GPNode(rootClass, self.pickALimit(rootClass))
nodesAtCurrentDepth.append(self.root)
depth = 0
# Initialize tree to one level short of depth
while depth < self.maxDepth - 2:
children = []
for node in nodesAtCurrentDepth:
children += self.giveChildren(node)
nodesAtCurrentDepth = children
depth += 1
# Add the last level of nodes, which should be terminating nodes (i.e. the possible predictions)
self.addTerminatingNodes(nodesAtCurrentDepth)
def initUnevenTree(self):
"""Creates a random decision tree that is not necessarily symmetrical or full"""
rootClass = self.pickAnAttribute() # Not the classToPredict
rootValue = self.pickALimit(rootClass)
self.root = GPNode(rootClass, rootValue)
nodesNeedingChildren = []
nodesNeedingChildren.append(self.root)
depth = 0
while depth < self.maxDepth - 2:
allChildren = []
for node in nodesNeedingChildren:
if node.attribute != self.classToPredict:
# Not a terminal node
allChildren += self.giveChildren(
node, includeClassToPredict=True)
nodesNeedingChildren = allChildren
depth += 1
self.removeTerminalNodes(nodesNeedingChildren)
self.addTerminatingNodes(nodesNeedingChildren)
def removeTerminalNodes(self, listOfNodes):
"""Removes all terminal nodes from <listOfNodes>"""
nodesToRemove = []
for x in listOfNodes:
if x.attribute == self.classToPredict:
nodesToRemove.append(x)
for node in nodesToRemove:
listOfNodes.remove(node)
def giveChildren(self, node, includeClassToPredict=False):
"""Gives children to <node>, returns these children"""
leftClass = self.pickAnAttribute(includeClassToPredict)
rightClass = self.pickAnAttribute(includeClassToPredict)
leftNode = GPNode(leftClass, self.pickALimit(leftClass))
rightNode = GPNode(rightClass, self.pickALimit(rightClass))
node.setLeftChild(leftNode)
node.setRightChild(rightNode)
return [leftNode, rightNode]
def addTerminatingNodes(self, currentNodes):
"""Adds terminal nodes to all nodes in <currentNodes>"""
# BE CAREFUL HERE; do I want the final split to have the possibility of predicting the same finalValue?
terminatingOptions = self.categoryLimits[self.classToPredict]
for node in currentNodes:
optionLeft = random.choice(terminatingOptions)
terminatingOptions.remove(optionLeft)
optionRight = random.choice(terminatingOptions)
node.setLeftChild(GPNode(self.classToPredict, optionLeft))
node.setRightChild(GPNode(self.classToPredict, optionRight))
terminatingOptions.append(optionLeft)
def pickAnAttribute(self, includeClassToPredict=False):
"""Picks a random attribte (each entry in allData contains every attribute as its keys)"""
if includeClassToPredict:
attr = random.choice(self.allCategories)
else:
attr = random.choice(self.categories)
return attr
def pickALimit(self, attribute):
"""Returns a random limit for the given <attribute> from somewhere in its specified values (as maintained by categoryLimits)"""
possibleLimits = self.categoryLimits[attribute]
try:
# Get random number between lower and upper limit
limit = random.uniform(possibleLimits[0], possibleLimits[1])
except TypeError:
# possibleLimits actually has strings, so <attribute> is categorical; return a random choice from the possibleLimits
limit = random.choice(possibleLimits)
return limit
def notTerminatingNode(self, node):
"""Determines whether or not <node> is a terminating node"""
if node.attribute == self.classToPredict:
return False
else:
return True
def evaluate(self, entries):
"""Tests all entries on his decision tree, returns the number classified correctly, and total entries attempted"""
correct = 0
total = len(entries)
for ent in entries:
if self.root.evaluate(ent):
correct += 1
return correct, total
def __str__(self):
"""Root has everything stored"""
return str(self.root)
